What Matters to an LLM? Behavioral and Computational Evidences from Summarization
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912866045526016 |
|---|---|
| author | Zhou, Yongxin Wu, Changshun Mulhem, Philippe Schwab, Didier Peyrard, Maxime |
| author_facet | Zhou, Yongxin Wu, Changshun Mulhem, Philippe Schwab, Didier Peyrard, Maxime |
| contents | Large Language Models (LLMs) are now state-of-the-art at summarization, yet the internal notion of importance that drives their information selections remains hidden. We propose to investigate this by combining behavioral and computational analyses. Behaviorally, we generate a series of length-controlled summaries for each document and derive empirical importance distributions based on how often each information unit is selected. These reveal that LLMs converge on consistent importance patterns, sharply different from pre-LLM baselines, and that LLMs cluster more by family than by size. Computationally, we identify that certain attention heads align well with empirical importance distributions, and that middle-to-late layers are strongly predictive of importance. Together, these results provide initial insights into what LLMs prioritize in summarization and how this priority is internally represented, opening a path toward interpreting and ultimately controlling information selection in these models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_00459 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | What Matters to an LLM? Behavioral and Computational Evidences from Summarization Zhou, Yongxin Wu, Changshun Mulhem, Philippe Schwab, Didier Peyrard, Maxime Computation and Language Large Language Models (LLMs) are now state-of-the-art at summarization, yet the internal notion of importance that drives their information selections remains hidden. We propose to investigate this by combining behavioral and computational analyses. Behaviorally, we generate a series of length-controlled summaries for each document and derive empirical importance distributions based on how often each information unit is selected. These reveal that LLMs converge on consistent importance patterns, sharply different from pre-LLM baselines, and that LLMs cluster more by family than by size. Computationally, we identify that certain attention heads align well with empirical importance distributions, and that middle-to-late layers are strongly predictive of importance. Together, these results provide initial insights into what LLMs prioritize in summarization and how this priority is internally represented, opening a path toward interpreting and ultimately controlling information selection in these models. |
| title | What Matters to an LLM? Behavioral and Computational Evidences from Summarization |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2602.00459 |